EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Lobsters·Aug 1·5 min read
LLMs won't break symmetric crypto
TL;DR
Anthropic's LLM Claude Mythos has revealed new cryptanalytic attacks, notably reducing the security level of the HAWK signature scheme and improving key-recovery methods for AES-128. While these findings are significant, they pose no immediate threat to established symmetric cryptography.
✦ Why It Matters
Consider standardizing your cryptographic frameworks to align with emerging research insights.
Key Takeaways
- Explore LLMs for cryptanalysis of post-quantum candidates.
- Understand that established symmetric algorithms are robust against LLM attacks.
- Utilize LLMs to identify potential errors in cryptographic proofs.
- Encourage unified frameworks in symmetric cryptanalysis research.
Full Summary
Anthropic's Claude Mythos has uncovered new cryptanalytic attacks, including a key-recovery attack on the HAWK signature scheme, reducing its security level from 128 bits to as low as 81 bits. Additionally, it improved a key-recovery attack on 7-round AES-128, although this poses no real threat to the full 10-round version.
Despite these findings, the article asserts that LLMs are unlikely to break established symmetric cryptographic algorithms due to their complex structures and extensive prior scrutiny. Symmetric ciphers are designed to eliminate exploitable patterns, making them resistant to new attack methods.
Furthermore, the empirical nature of differential cryptanalysis, which relies on statistical biases, complicates the ability of LLMs to discover new vulnerabilities. Overall, while LLMs can assist in cryptographic research, their impact on breaking established symmetric schemes is expected to be minimal.